Key Moments
Big Tech Ran Out Of Ideas — And AI Is The Cover Story — We Had To React
Impact TheoryWant to know something specific about what's covered?
We've already dissected every moment. Ask and we will deliver (with timestamps).
Key Moments
Big Tech's AI investments are a massive financial black hole; Open AI alone lost $20.9B in 2025, proving the industry is overhyped and lacks a sustainable business model.
Key Insights
OpenAI burned $20.9 billion in 2025, with costs increasing linearly with revenue and no proof of margin improvement, according to FTI-reported auditive financials.
Alex Karp, Palantir CEO, argues enterprises are wary of AI due to token costs, IP risks, and the fact that AI companies train on proprietary data and then launch competitive products.
The current business model of AI companies is characterized by encouraging waste, not charging on outcomes, and inherently hallucination-prone models, leading to significant financial losses.
Revolutionary technologies often bankrupt their first generation of investors with massive infrastructure build-outs, a pattern expected to repeat with AI, benefiting a later 'inheritance generation'.
Global data center construction is projected to require trillions of dollars, with the debt markets being tapped out, evidenced by Google's $85 billion equity raise.
Tech companies like Microsoft, Google, and Meta are conflating AI losses with their overall growth; they do not disclose AI-specific revenues because they are losing money on AI initiatives.
The staggering financial drain of generative AI
The current state of the AI industry is marked by extraordinarily high costs and questionable financial viability, a sentiment echoed by bear case proponents like Ed Zitron. OpenAI, for instance, reportedly burned through an astounding $20.9 billion in 2025, according to FTI-reported financials. A critical issue highlighted is the linear increase in costs that mirrors revenue growth, with no demonstrable path to improving profit margins. The argument is that even advanced technologies like specialized silicon or theoretical breakthroughs won't significantly reduce these operational expenses. This fundamental mismatch between significant financial losses and the revenue generated by these AI companies is a core concern, suggesting a potential bubble. For instance, Palantir CEO Alex Karp points out that enterprises are hesitant to fully adopt AI due to exorbitant token costs and concerns about intellectual property being compromised. AI companies often train on customer data and then leverage that knowledge to create competing products, a practice that erodes trust and discourages long-term investment from businesses.
Why enterprises are wary of current AI offerings
Alex Karp, CEO of Palantir, articulates a significant point of friction for enterprise adoption of AI: the inherent risks and lack of tangible value for businesses. He notes that enterprises are reluctant to engage with AI due to the "token costs" and the potential loss of their intellectual property (IP). A key concern is that AI companies train their models on the proprietary data provided by these enterprises. Subsequently, these AI companies can then use this trained knowledge to develop their own competing products or services, effectively turning a client's data against them. This practice has been observed in the market, leading to a general sentiment of distrust. Karp suggests that a solution lies in creating an "obfuscation layer" where companies can securely feed their data into AI models without the fear of it being stored, retained, or used to mimic them. This layer would allow for customization and the creation of proprietary AI output, making the technology truly valuable to the individual enterprise.
The problematic business model of AI companies
A critical observation made about the current AI industry is how many companies are structured to "encourage waste" rather than deliver measurable outcomes. Unlike traditional businesses that charge for success or results, AI companies, especially those dealing with large language models (LLMs), cannot reliably charge for outcomes because their models are prone to 'hallucinations'—generating incorrect or nonsensical information. This inherent unreliability leads them to charge based on usage or tokens, incentivizing customers to spend more money. Furthermore, there's a pattern of AI companies attempting to "pill for your ideas," meaning they absorb information from user interactions and then develop competitive offerings. This strategy, coupled with the fact that many of these AI-generated products are not particularly superior to existing solutions, raises serious questions about the sustainability of their business models. The founders themselves often admit they don't fully know what can be built with their technology, relying on external innovation while keeping costs high.
AI as a commodity and the need for proprietary solutions
The long-term future of AI, according to the discussion, likely hinges on intelligence becoming a commodity, similar to electricity. The current model, where companies charge for raw AI capabilities, is seen as unsustainable. Alex Karp's emphasis on creating a "middle layer" where businesses can develop unique, proprietary AI applications is crucial. This approach would allow companies to tailor AI to their specific needs, generating custom outputs and maintaining a competitive advantage. Without this move towards proprietary applications, AI risks becoming a generic utility that offers little differentiation. The analogy is drawn to the internet's early days: while the infrastructure was built, it was the subsequent development of unique applications like the iPhone and Uber that unlocked its true potential and created massive value. The success of AI will likely depend on businesses finding ways to make it uniquely theirs, rather than relying on generic, interchangeable models.
The historical pattern of revolutionary technology investors
History provides a stark warning for those investing in revolutionary technologies with massive infrastructure build-outs: the first generation of investors often faces bankruptcy. This phenomenon is attributed to the immense upfront costs required to develop the foundational infrastructure. For example, the development of railroads or fiber optic networks involved huge capital expenditures that initially overwhelmed the companies involved. The profitable phase typically comes later, with the 'inheritance generation' of investors who build their businesses on top of the existing, already-paid-for infrastructure. They benefit from the groundwork laid by the pioneers without bearing the initial debt burden. This pattern is highly likely to repeat with AI, suggesting that early investors in AI infrastructure and development may not see returns, while subsequent companies leveraging that infrastructure could thrive. The sheer scale of debt accumulation in the AI sector mirrors these historical precedents, raising concerns about financial stability.
The debt trap and potential market collapse
A significant concern is the immense debt being accumulated to fund the AI boom, creating a potential risk for the broader economy. The massive capital expenditure (capex) required for data centers and AI development is being financed through debt markets, which are increasingly tapped out. This situation is compared to the 2008 financial crisis, where debt was hidden and diversified across various financial instruments to mitigate risk. Banks are expected to diversify their exposure, potentially embedding this AI-related debt into insurance, index funds, and retirement programs. There's a risk that this debt could be packaged and assigned high credit ratings, masking its true risk. The script suggests that the first hyperscaler to pull back on capex will be rewarded by markets, and any financing failures for major AI companies would be a significant signal. The ultimate trigger for a market reversal could be when data center debt stops being issued, indicating a significant contraction in available capital for expansion.
AI's role beyond LLMs and the democratization of intelligence
While LLMs are currently the focus, the broader trajectory of AI is about the democratization of intelligence, making it ubiquitous. The argument is that intelligence will be integrated into virtually every object, transforming everyday items into 'smart' devices—from chairs to televisions. This vision suggests a future where AI is not just a tool for complex tasks but a fundamental layer of enhancement for countless applications. Even if LLMs themselves reach a plateau in terms of raw intelligence, the ongoing integration and optimization of existing AI capabilities will continue to drive transformation. The medical field is cited as an example, where AI's ability to analyze massive datasets, such as protein folding, is already leading to breakthroughs in drug discovery and treatment, demonstrating the power of pattern recognition beyond immediate revenue generation. This pervasive integration implies a long runway for AI's impact, even if the current LLM business models are flawed.
The arms race and governmental influence
Beyond financial concerns, the development of advanced AI introduces a critical national security and geopolitical dimension, akin to an arms race. The existence of powerful AI models, such as the rumored 'Fable 5,' capable of sophisticated hacking and exploitation of legacy systems, has prompted governmental intervention. When companies like Anthropic report such capabilities, it triggers serious national security discussions, with governments potentially restricting access to prevent adversaries from acquiring such technology. The illicit sale of AI technology to countries like China, even through intermediaries, highlights the complex regulatory landscape. Governments may find themselves in a position where they need to support and control AI development due to its strategic importance, creating a dynamic where national interest could override market forces or pure business viability. This 'weapon system' aspect of AI adds another layer of complexity to its future development and deployment.
Mentioned in This Episode
●Products
●Software & Apps
●Companies
●People Referenced
Common Questions
AI companies face unsustainable costs because their expenses, especially for infrastructure and compute, increase linearly with revenue. This makes it difficult to improve margins, as demonstrated by OpenAI's significant losses.
Topics
Mentioned in this video
A commentator described as a 'bear' on AI, who believes the industry is overhyped and unsustainable financially.
Mentioned in the context of specialist silicon that is not expected to bring down AI costs.
CEO of Palantir, whose comments on enterprises being wary of AI due to data IP risks and high token costs are discussed.
Mentioned as being shocked by Anthropic's attempts to compete with Figma.
CEO of Anthropic, mentioned in the context of AI companies not knowing what users will build with their technology.
CEO of OpenAI, mentioned regarding AI companies not knowing what users will build and encouraging innovation through others.
Discussed for his vision in robotics and AI, his perspective on hyperscalers leasing out compute power, and his involvement with X AI.
Mentioned in the context of Oracle's potential financial jeopardy if OpenAI fails to pay for data center services.
Discussed for its significant financial losses, burning $20.9 billion in 2025, and the unsustainable cost model of large language models.
Has created a 'middle layer' called Ontology to address enterprise concerns about data privacy and token costs in AI.
Mentioned as a company against which AI companies have launched competitive products after training on their data.
Criticized for encouraging waste with LLMs, not charging for outcomes, and attempting to compete with companies like Figma using client data.
Cited as an example of a later, transformative application built upon initial internet infrastructure, similar to how AI might evolve.
A CRM system that Quo integrates with, mentioned during an advertisement for Quo.
An automation tool that Quo integrates with, mentioned during an advertisement for Quo.
Mentioned for its self-driving car technology, which leverages AI and requires extensive training data.
Mentioned in relation to Elon Musk's potential hedging bets and as a new AI company in the space.
Highlighted as a risky investment due to building massive data center capacity (7.1 GW) for OpenAI, with potential non-payment risks.
Discussed for experimenting with AI for video game apps and its potential role in revenue growth, though the speaker disputes AI as the primary driver.
Mentioned as the company whose GPUs are being rented back by data centers, indicating a lack of diverse demand in the industry.
Mentioned for raising $85 billion in equity and for its alleged involvement in selling technology to Chinese companies.
Mentioned as an AI model launch that caused a significant market value drop, spooking investors about the AI race between the US and China.
Palantir's product designed to act as a middle layer for enterprises using AI, addressing concerns about data usage and costs.
An Anthropic product that allegedly tried to replicate Figma, indicating a pattern of AI companies competing with their clients.
A business phone system mentioned as an advertiser, highlighting its AI agent for handling after-hours calls and integrating with other tools.
A powerful AI model that Anthropic found could hack systems, leading them to inform the government, and highlighting AI's potential as a weapon system.
More from Tom Bilyeu
View all 118 summaries
117 minApple Is Trying to Kill OpenAI Right Before Its IPO
153 minUS Hits Iranian Infrastructure. DSA Commits Treason? Spencer Pratt’s Bombshell Accusation
56 minChina Says They Shut Down Gold Trading To Protect You — That's Not Why — We Had To React
101 minOil Explodes As Ceasefire Fails, Japan Bond Crisis Threatens US Debt, & Platner Ditched By Dems
Ask anything from this episode.
Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.
Get Started Free